Founder, Paperwork Labs · San Francisco

Data platform engineer / full-stack AI builder.

I've spent my career turning operating problems into working systems—from bringing small businesses fully online to Cruise's commercial data warehouse, Credit Karma's recommendation and streaming pipelines, and full-stack AI products at Paperwork Labs. My work spans data architecture and governance, agent runtimes, fintech and compliance products, identity, and the delivery automation behind them.

Explore selected work Download résumé

Selected engineering evidence

repository commits: 3,700+
FastAPI codebases: 7
typed agent personas: 17
GitHub workflows: 54
tax jurisdictions: 51
migrated to Paperwork ID: 6 apps · 4 APIs

Selected systems

Six products, one operating substrate.

I built these as connected systems: shared identity, shared data doctrine, reusable backend foundations, and one agent runtime coordinating work across the portfolio.

Portfolio + signals

AxiomFolio

Portfolio intelligence and trading workflows that combine live positions, signals, market context, and risk-aware execution paths.

Next.js
FastAPI
PostgreSQL
Redis
MCP

Consumer tax

FileFree

Mobile-first tax preparation with W-2 and 1099 OCR, integer-cents calculations, a MeF document lifecycle, and state-tax coverage across all 50 states plus D.C.

Next.js
FastAPI
Python
GCP Vision
MeF

LLC formation

LaunchFree

LLC-formation workflow with deterministic state-portal handlers under certification, compliance dates, and a shared data layer for state rules and fees.

Next.js
FastAPI
Playwright
PostgreSQL
MCP

B2B compliance

Distill

B2B compliance product in development: a CPA-facing dashboard plus firm-scoped tax API routes for structured document workflows.

Next.js
FastAPI
Firm-scoped APIs
Paperwork ID

Company OS

Studio

Founder command center for products, agents, conversations, secrets, infrastructure, and operating decisions.

Next.js
TypeScript
Paperwork ID
Vercel

Agent runtime

Brain

Cross-product agent runtime for persona routing, memory, tool use, scheduled work, and dispatch.

Python
FastAPI
SQLAlchemy
PostgreSQL
Redis
MCP

Architecture under constraints

The systems behind the products.

Four projects show how I approach durable engineering: make policy executable, automate the path to production, migrate critical infrastructure without losing control of the rollback—and hold a data platform to the standard people bet daily operations on.

Agent runtime

Policy, memory, and tools—not another chat wrapper.

Brain is the orchestration layer behind the portfolio. It routes work through typed persona contracts, enforces model and cost policy before provider spend, stores episodic and procedural memory, schedules recurring work, and exposes product capabilities through MCP. Studio provides the human control plane.

FastAPI
SQLAlchemy
Neon
Upstash
MCP

17 typed persona contracts define model choice, escalation, cost ceilings, tool routing, and human-review flags.
Episodic and procedural memory plus provenance URIs make outputs traceable.
APScheduler, HTTP, and MCP connect recurring work, product tools, and the Studio control plane.

Delivery automation

A production path designed to drain its own queue safely.

The monorepo’s PR pipeline treats draft and ready state as a cost-control signal, gives PR Desk a capped mechanical-repair budget, binds AI review to the exact commit SHA, and merges agent work only when labels, reviews, checks, mergeability, and branch protection agree. Post-merge automation rebuilds derived indexes, while previews remain path-scoped.

GitHub Actions
Python
TypeScript
Hetzner
Vercel

54 versioned workflows cover CI, review, triage, repair, rebase, merge, deployment, smoke tests, and system health.
Review verdicts fail closed and are bound to the exact pull-request head SHA.
Draft mode limits expensive CI; ready-for-review intentionally turns the full gate set on.

Identity migration

A staged identity cutover across ten surfaces.

I replaced Clerk with self-hosted Better-Auth across six frontend apps and four FastAPI backends while preserving each product’s sign-in experience. Paperwork ID centralizes users and sessions at accounts.paperworklabs.com, exchanges short-lived one-time handoff tokens for apex-scoped EdDSA JWTs, and lets backends verify them through JWKS. The greenfield cutover landed in roughly 13 PRs with rollback points between waves.

Better-Auth
Next.js
EdDSA JWT
JWKS
PostgreSQL

One identity host serves product-specific sign-in shells across separate domains.
Single-use handoff tokens exchange for product-scoped sessions instead of sharing cross-apex cookies.
A staged migration plan preserved rollback points through frontend, backend, and package cutovers.

Data platforms · before Paperwork Labs

A commercial data warehouse 300 people trusted every day.

Before founding Paperwork Labs I built the data platforms behind two venture-scale products. At Cruise I designed Medallion—a Bronze/Silver/Gold layered architecture—as the single source of truth for 600+ users, then built the governance layer around it: a data catalog, automated quality pipelines, lineage tracking, and role-based access tiered by data sensitivity. At Credit Karma I built the email recommendation pipeline serving 120 million members and shipped the company's first streaming pipeline for event-driven engagement.

BigQuery
Dataflow
Apache Beam
Kafka
Snowflake
Airflow
Hex

Medallion's Bronze/Silver/Gold layers cut data inconsistencies 80% across 600+ users of Cruise's Commercial Data Warehouse (300+ daily active users).
A governance framework—catalog, automated quality checks, lineage, tiered RBAC—caught 95% of anomalies before they reached production.
Credit Karma's email recommendation system reached 120M+ members and contributed to $20M+ in incremental monthly revenue; a companion pipeline was the company's first streaming system.

Before Paperwork Labs

From small-business delivery to data platforms and AI systems.

The through-line is ownership: understand the operating problem, build the platform around it, and make the result usable by the people doing the work.

members served by Credit Karma's email recommendation system: 120M+
incremental revenue the recommendation system contributed to: $20M+/mo
daily active users of Cruise's Commercial Data Warehouse: 300+
fewer data inconsistencies via the Medallion architecture: 80%
of data anomalies caught before reaching production: 95%
engineers on Cruise's Commercial Data team as it scaled: 2 → 15

Career figures are résumé-sourced and self-reported, unlike the repo-derived engineering evidence above.

Cruise
2021–2026
Senior Data Engineer II
Architected Medallion, Cruise's Bronze/Silver/Gold data platform, and the governance framework around it—catalog, automated quality checks, lineage, tiered RBAC—that kept 95% of anomalies out of production. Maintained the Commercial Data Warehouse for 300+ daily users across finance, legal, and customer operations as the team grew from 2 to 15 engineers.
Credit Karma
2019–2021
Senior Data Engineer
Built the email recommendation pipeline for 120M+ members, contributing to $20M+ in incremental monthly revenue. Shipped Credit Karma's first streaming pipeline for event-driven engagement and stood up GDPR-compliant pipelines for the 20M-member UK market.
Lithium Technologies
2017–2019
Full-Stack Engineer
Maintained LIA, a community platform serving 800M members across 500+ communities—4B page views and 7B API calls a year. Broke the monolith into server-rendered microservices for the Products V2 launch.
Alpha Technologies
2014–2017
Senior Software Engineer
Modernized a J2EE device-monitoring platform used across 80% of the company's broadband customer base—Java 6→8, Spring, Oracle 12c, and an Angular front end that cut load times from 2 minutes to 5 seconds. Led a 5-person team.
Small-business digital studio
2010–2014
Co-founder / Engineer
Helped owner-operated businesses establish a complete internet presence, translating plain-language needs into shipped websites and practical digital workflows.